Analysis of Crevasse Patterns as Indicators of Ice Dynamics Using Structural Glaciology and Geostatistical Classification

Crevasse patterns are the writings in a glacier's history book --- the movement, strain and deformation frozen in ice. Therefore by analysis of crevasse patterns we can learn about the ice-dynamic processes which the glacier has experienced. Direct measurement of ice movement and deformation is time-consuming and costly, in particular for large glaciers; typically, observations are lacking when sudden changes occur. Analysis of crevasse patterns provides a means to reconstruct past and ongoing deformation processes quantitatively. Crevasse patterns are utilized as a source of geophysical information. Our structural glaciology approach builds on methods adapted from structural geology and continuum mechanics. In slow-moving ice, ductile deformation prevails, and the related processes are largely understood. The dynamics of fast-moving glaciers are much less well understood but manifest themselves in the formation of crevasses. The crevasses result from brittle deformation and are considered cracks in the continuum. Geostatistical crevasse pattern analysis is a subset of geostatistical surface classification, a method designed on the background of the theory of regionalized variables. In a generalization of the well-known variogram, the structure function most commonly used in geostatistics, first- and higher-order vario functions, experimental ``ordinary" and residual vario functions are defined. Crevasse patterns are organized in classes according to principles of structural geology, examples of classes are one-directional, two-directional, rhombic, en-échelon, chaos (geometric classification), extensional, compressional, and shear types (kinematic classification). Parameters extracted from generalized vario functions are combined into feature vectors, and it can be shown that specific feature vectors are characteristic of each of the crevasse classes. Association between geostatistic parameters and feature vectors and crevasse patterns is achieved using deterministic decision rules or neural networks, and the reconstruction of the dynamics is then clear from the structural glaciology design of the crevasse classes. Following a classification of fast-moving ice, we explore data from Jakobshavn Isbr\ae , Greenland, prototype of a continuously fast-moving glacier (autonomous system, type A1), and Bering Glacier, Alaska, prototype of a surging glacier (discrete deformation events, type B1). In this typology, more complex systems are defined by (B2) sequences of events as recorded by multiphase crevasse patterns (several surge stages) and (A2) time-dependent dynamic processes resulting in crevasse patterns.